Notes: Perplexity's Race to Build Agentic Search

This post summarizes Aravind Srinivas’s talk at Y Combinator (July 2025) on Perplexity’s strategy, competition, and the future of AI-powered search.

The Big Bet: The Browser

Perplexity’s next major bet is building a browser called Comet. The vision is to create a “cognitive operating system” rather than just another browser.

Key features of the browser vision:

• One omni-box for navigation, informational queries, and agentic tasks
• AI assistant available on every web page as a sidecar
• Launch multiple tasks in parallel running asynchronously
• Integration with personal data: contacts, email, calendar, Amazon, social media
• Research capabilities for real estate, markets, and more

Why a browser?

• Chrome was exciting when each tab was its own process
• Now imagine each query or prompt being its own process
• The browser enables things chatbots cannot: accessing other tabs, browsing history, completing forms, paying credit cards, buying things, doing periodic recurring tasks

Competition and the Innovator’s Dilemma

Aravind acknowledges the competitive landscape but sees structural advantages for startups:

The startup advantage:

• “If something is really worth doing, it’s only natural that people with a lot of funding will go and do it”
• Perplexity was a great product, now everyone is trying to build something that can answer any question with sources
• Cursor was a great product, now OpenAI is trying to buy its competitor, Anthropic launched Claude Code, Google has its own rival tool

Google’s innovator’s dilemma:

• Google had the 4th or 5th best models for much of 2023-2024
• As a startup outside Google, you had access to AI that was better than what Google internally had - unprecedented
• Google can’t give direct answers because it would hurt their ad revenue from Booking, Expedia, Kayak, Amazon, Walmart
• One mistake tanks their stock (remember the Bard demo failure that dropped stock 6%)
• Startups can make mistakes and it’s fine

Why Google struggles:

• Much bigger business, not founder-led anymore
• Hard to take risks
• Great people, but incentive structure makes it hard to do long-term correct things
• Same feature launched year after year with different names, different VPs, different groups - but never launched to everybody

The Perplexity Origin Story

Starting without a clear idea:

• Started the company without actually having a clear idea of what to build (opposite of YC advice)
• In AI’s fast-moving landscape, you don’t have to rigidly stick to any one idea
• But don’t change the idea every week either

The pivot from SQL to search:

• First tool was natural language SQL - searching over relational databases
• Built Twitter search using language models, organizing Twitter data into relational tables
• Realized it’s better to scale across the web with unstructured data
• Bet on language models doing all the reasoning and parsing

Early validation:

• Launched 7 days after ChatGPT launch (when ChatGPT didn’t have web search)
• New Year’s Eve saw close to 700,000 queries
• Despite: crappy name, slow (7 seconds per query), lots of hallucinations, no-name company, $1-2M seed funding
• People were sharing screenshots on New Year’s Eve instead of watching Netflix

The aha moment:

• When Google announced Bard, everyone said “Bard will do whatever you’re doing”
• Realized Google has to build a separate product but can never capitalize on their core distribution
• The thought: “Why do I have to build Bard? Why not just do it on Google where you have all the distribution?”

Running the Company

Team size: About 200 employees

AI coding tools adoption:

• Made it mandatory to use at least one AI coding tool
• Mix of Cursor and GitHub Copilot internally
• ML researchers upload screenshots of pseudo code from papers, ask Cursor to implement algorithms
• Experimentation time reduced from 3-4 days to literally 1 hour
• CEO gives feedback via screenshots with arrows, engineers upload to Cursor to make Swift UI changes

Vibe coding caveats:

• Don’t want to vibe code everything
• Don’t want a vibe coder fixing production infra issues
• Need people well-trained in regular software engineering, infrastructure, distributed systems
• Front-end and design is where tremendous adoption is happening
• AI tools introduce new bugs, and many people don’t know how to fix them

CEO involvement:

• Aravind loves triaging and fixing bugs personally
• “Is that the best use of the time of a CEO?” - Many would think otherwise
• Even Sundar is doing bug support on X now

Competing with OpenAI

ChatGPT is the most successful consumer AI product. Competing against it is very difficult.

The browser strategy:

• Comet browser will be an abstraction layer above chatbots
• Could feed all your ChatGPT chats into the browser AI
• Don’t have to worry about memory or personalization
• Access other tabs, browsing history, complete forms, pay credit cards, buy stuff, do research, periodic recurring tasks

What makes the browser hard to copy:

• Much harder to copy than yet another chat tool
• Building mobile versions will take many months of engineering
• Not worried about someone else copying - it’s going to take time for anybody

Enduring Value and Brand

What protects AI companies:

• Brand has significant value once you acquire millions of paying users
• You earn the right to survive and keep building
• Narrative is very important to brand - communicate why you need to exist
• For Perplexity: focus on accuracy, speed (time to first token), answer presentation

Network effects in AI:

• Brand has network effects (people tell each other)
• But no AI product has within-app network effects like WhatsApp
• You can easily export ChatGPT history and upload elsewhere
• AI doesn’t have the lock-in that messaging apps have

Key Quotes

“You have to innovate. You have to move faster than everybody else. And it’s like running a marathon but at an extremely high velocity.”

“The only mode you have is speed.”

“Google IO last year was AI overview and perplexity is dead. This year was AI mode and perplexity is dead. I read all of that too and it’s always fun. I love it actually.”

“If startups can succeed then it’s going to be monopolies getting bigger and bigger and that’s not great for the world.”